Google DeepMind and Singapore develop on-device AI scam detector
Tech

Google DeepMind and Singapore develop on-device AI scam detector

TechNews Editorial
TechNews EditorialOct 14, 2026 · 1 min read
Share

Why it matters

Google DeepMind says the model could warn users about manipulative scam conversations while processing their chats on their own devices.

The facts

  • Google DeepMind and GovTech Singapore developed an on-device model to detect scam attempts in conversations.
  • The model uses a two-stage check and warns users before they share sensitive information or money.
  • Its rollout in applications will be phased, and conversation access will require user consent.

Advertisement

Google DeepMind and Singapore’s Government Technology Agency, GovTech Singapore, have developed an AI model designed to spot scam attempts in conversations on a user’s device. The model will warn users when it detects manipulative patterns. Its rollout in applications will be phased, the company says.

Scams account for more than half of Singapore’s reported crimes and have caused more than S$4 billion in losses since 2019, according to Google DeepMind. The company says scams built around convincing conversations now account for up to 80% of losses.

The model checks conversations in two stages

The partners fine-tuned Google’s lightweight Gemma model using local scam conversations, including impersonations of public institutions, banks and local figures. Google DeepMind says the work is intended to help the model distinguish ordinary conversation from manipulation.

A low-power classifier first screens incoming messages for possible threats. If it flags one, the fine-tuned Gemma model examines a larger part of the conversation to confirm the threat or rule out a false alarm. A detected pattern triggers a warning on the user’s device advising them to pause before sharing sensitive information or money.

A man sets his bank card aside and withdraws his hand from a phone, pausing before sharing financial information.
Illustration: AI & Tech News

Conversation analysis stays on the device

Google DeepMind says the model processes conversations on the device rather than sending them to a cloud server for scanning. Once deployed, it will access a user’s on-device conversations only with consent and under the application’s terms of use. The company says neither Google nor the Singapore Government will be able to access those conversations through the model’s deployment.

Fine-tuning took place in an environment managed by GovTech Singapore. According to Google DeepMind, the training excluded private and personally identifiable citizen information and used anonymised data from scam reports and intelligence feeds. The partners also tested the model against synthetic dialogues and local datasets.

The collaboration was presented at Singapore International Cyber Week 2026. Google DeepMind says the next step is a phased rollout of the model within applications, developed with the Singapore Government.

Source: Google DeepMind

Newsletter

Get the best AI & tech news daily

A concise daily digest. Unsubscribe anytime.

We use your email only to send this newsletter.

Keep reading